NEX or no NEX? A high angular resolution diffusion imaging study

INTRODUCTION: To remove noise artifacts and enhance SNR, diffusion-weighted (DW) measurements are often repeated multiple times (NEX) along each gradient direction (typically, from 2 to 4 measurements) and each DW image is averaged. This averaging increases acquisition time considerably. It is also not clear what effects it has on the reconstructed diffusion profiles such as the orientation distribution function (ODF) [1] or fiber density orientation (FOD) function [2], also called fiber ODF (fODF) [3]. In magnitude MRI, noise is Rician-distributed [4], and is often conveniently approximated as Gaussian in high SNR regions. In the context of diffusion tensor imaging (DTI), recent attempts have been made to account for the Rician noise to regularize the DW data [5], to estimate the diffusion tensor [6], or to perform both tasks simultaneously [7]. However, among the existing methods to estimate or regularize ODF reconstructions from high angular resolution diffusion imaging (HARDI), the Rician noise bias has just started to be addressed [8,9,10,11], but the impact on the reconstructed HARDI profiles such as the ODF and fODF have not been validated yet on real data. In particular, it is important to make sure that angular resolution of HARDI is not lost when removing Rician noise. In this work, the DW images are filtered with a specific Rician-based, structure-preserving, Non-Local Means method (NLMr) [5], which recently showed to outperform other state-of-the-art methods to denoise DTI at low b-values. The filtering is applied on the raw DW images before estimating the HARDI profiles such as the ODF and the fODF. We show that multiple DW measurements and averaging can be avoided because NLMr filtering of the individual DW images can remove the Rician noise bias while preserving angular resolution. The method also produces better quality generalized fractional anisotropy (GFA) maps and more coherent ODF fields.